基于LSTM的小井眼阵列感应测井响应预测

Prediction of Responses of Slim Hole Array Induction Logging Based on LSTM Network

快速准确地获取测井响应曲线是进行后续测井反演的基础。为了对测井响应进行高效的正演建模,将LSTM神经网络应用于小井眼阵列感应测井响应的预测中。数据集来源于COMSOL软件建模仿真得到的小井眼感应测井响应数据。在离线训练阶段,LSTM模型从测井响应数据库中提取到小井眼感应测井响应的数据特征。给定简单各向同性层状地层和复杂各向同性地层两种地层参数,训练得到的模型可用于预测阵列感应仪器的小井眼测井响应。研究表明,LSTM预测结果与基于测井响应数据库传统多维插值法的结果基本吻合。在保证预测结果精度的前提下,LSTM预测极大地提高了不同地层场景下测井响应的预测时间,且具有良好的泛化能力。

Fast and accurate acquisition of logging response curves is the basis of subsequent logging inversion.In order to efficiently model the logging response,LSTM is applied to predict the response of small-hole array induction logging.The data set is derived from COMSOL software modeling and simulation of the response data of small-hole induction logging.During the offline training phase,the LSTM model extracted data features from the response of the logging dataset for the induction logging response in the slim-hole.Given two kinds of formation parameters,simple i-sotropic stratiform and complex isotropic stratiform,the trained model can be used to predict the small-hole logging response of the array induction instrument.The results show that the predicted response of LSTM is close to the real logging response data based on multidimensional interpolation.On the premise of ensuring the accuracy of prediction results,this method greatly improves the prediction time of logging responses under different formation scenarios and has good generalization ability.